Cloud computing cabinet power distribution method and equipment based on AI optimization and medium

Abstract

The invention discloses a cloud computing cabinet power distribution method and device based on AI optimization and a medium, and relates to the technical field of computer power management, and the method comprises the steps: constructing a power demand prediction feature set according to an electric power state sequence, executing forward reasoning through a trained recurrent neural network model, and generating a power demand prediction value set; performing power supply capability constraint verification on the power demand predicted value set based on a cabinet power supply connection model, and performing evaluation and screening in combination with a power matching degree, power fluctuation and a power supply safety margin to form a power supply power distribution scheme selection set; and mapping the power supply power distribution scheme selected set into a power supply distribution parameter set based on the port-path mapping relationship, and performing state updating in combination with power deviation evaluation to generate a cloud computing cabinet power distribution state set. The effect of guaranteeing power supply safety and stability in a complex industrial cloud computing environment is achieved, and the energy efficiency utilization rate is improved while power supply safety and stability are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of computer power management technology, and in particular to a cloud computer cabinet power distribution method, device and medium based on AI optimization. Background Technology

[0002] In the field of industrial cloud computing, power distribution for cloud computing cabinets is one of the key technologies for ensuring the stable operation of computing equipment. Conventional power distribution methods are usually based on the cabinet's power supply topology information, employing static configuration or rule-driven dynamic adjustment strategies. These methods make basic power distribution decisions by collecting equipment power data and combining it with predefined power interface connection relationships. For example, conventional methods may rely on hardware monitoring modules to monitor current and voltage values ​​in real time and trigger distribution adjustments based on historical power averages or fixed thresholds. Such methods can maintain basic power supply balance in environments with relatively stable loads.

[0003] However, conventional methods have certain limitations in power demand forecasting and power allocation assessment. On the one hand, in the power demand forecasting stage, conventional methods often employ time series analysis techniques such as moving averages or linear regression, which struggle to effectively capture the nonlinearity and time-series dependence of load fluctuations in industrial cloud computing environments, leading to significant prediction errors. On the other hand, in selecting power allocation schemes, conventional methods often focus on a single indicator of power matching degree, neglecting the multi-dimensional impact of power fluctuation characteristics and power supply safety margins, resulting in insufficient adaptability of allocation schemes in dynamic scenarios. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an AI-optimized power distribution method for cloud computer cabinets to solve the problems of low intelligence level and inaccurate power prediction in cloud computer cabinet power distribution.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an AI-optimized power distribution method for cloud computer cabinets, comprising,

[0008] Based on the power supply topology information of the cloud computer cabinet, the power supply interface, interface connection relationship and power supply path are analyzed in layers and a hierarchical reference relationship is established to construct the cabinet power supply connection model.

[0009] Extract the internal power management object of the device corresponding to the power supply endpoint from the cabinet power connection model, and perform synchronous sampling and power calculation to generate an electrical power state sequence;

[0010] A power demand prediction feature set is constructed based on the power state sequence, and forward inference is performed through a trained recurrent neural network model to generate a set of power demand prediction values.

[0011] Based on the cabinet power connection model, the power demand prediction set is checked for power capacity constraints, and the power matching degree, power fluctuation and power supply safety margin are evaluated and screened to form a set of selected power allocation schemes.

[0012] Based on the port-path mapping relationship, the selected set of power allocation schemes is mapped to the set of power allocation parameters, and combined with the power deviation assessment to perform status updates, a set of power allocation statuses for the cloud computer cabinet is generated.

[0013] As a preferred embodiment of the AI-optimized cloud computing cabinet power distribution method of the present invention, the step of performing hierarchical analysis of power supply interfaces, interface connection relationships, and power supply paths based on the power supply topology information of the cloud computing cabinet is as follows:

[0014] Based on the power supply topology information of the cloud computer cabinet, the power supply structure of the cloud computer cabinet is abstracted at the node level based on the power network topology model. Interface identifiers are assigned to interfaces with power supply start-point and power supply end-point attributes, and a power supply interface layer interface set is generated.

[0015] The power supply topology information for the cloud computer cabinet includes the interface number, interface type, starting and ending interface numbers of the cable connection, and interface direction attribute.

[0016] Using the power supply interface layer interface set as the node set in the power network topology model, and combining the power supply topology information of the cloud computer cabinet, the connection endpoints are paired and the connection attributes are recorded to generate a power supply connection layer connection relationship set.

[0017] Based on the interface set of the power supply interface layer, the interface identifiers are linked in a chain according to the path reachability rules in the power network topology model along the connection order of the power supply connection layer connection relationship set to generate the path relationship set of the power supply path layer.

[0018] As a preferred embodiment of the AI-optimized cloud computer cabinet power distribution method described in this invention, the steps for establishing hierarchical reference relationships and constructing a cabinet power connection model are as follows:

[0019] Extract power supply path identifiers from the power supply path layer path relationship set, establish a corresponding relationship with interface identifiers, and generate port-path mapping relationship;

[0020] Based on the interface set of the power supply interface layer, the connection relationship set of the power supply connection layer, and the path relationship set of the power supply path layer, a hierarchical reference relationship between interface identifier, connection endpoint, and path link is established through hierarchical topology modeling to construct a cabinet power supply connection model.

[0021] As a preferred embodiment of the AI-optimized cloud computer cabinet power distribution method described in this invention, the steps of extracting the internal power management object of the device corresponding to the power supply endpoint from the cabinet power connection model, and performing synchronous sampling and power calculation to generate an electrical power state sequence are as follows.

[0022] Extract interface identifiers with power supply endpoint attributes from the cabinet power connection model and map them to a set of power management objects inside the device.

[0023] Synchronous sampling is performed on the set of power management objects inside the device, and timestamp alignment, abnormal sampling point removal and sampling consistency verification are performed to generate power supply sampling data;

[0024] Power values ​​are obtained from power supply sampling data and integrated into an electrical power state sequence.

[0025] As a preferred embodiment of the AI-optimized cloud computer cabinet power distribution method of the present invention, the steps for constructing a power demand prediction feature set based on the power state sequence are as follows:

[0026] The power state sequence is segmented, and the power change amplitude and power change trend features are extracted.

[0027] The feature combination processing is performed on the change magnitude feature and the change trend feature, and then normalization processing is performed to generate a power demand prediction feature set.

[0028] As a preferred embodiment of the AI-optimized cloud computer cabinet power distribution method described in this invention, the step of generating a set of predicted power demand values ​​by performing forward inference through a trained recurrent neural network model is as follows:

[0029] Using historical power state sequences as training data, the parameters of the recurrent neural network model are iteratively trained by minimizing the mean square error loss function to obtain the trained recurrent neural network model.

[0030] The power state sequence is input into the trained recurrent neural network model, and the power demand for future time windows is predicted through forward inference, generating a set of predicted power demand values.

[0031] As a preferred embodiment of the AI-optimized cloud computer cabinet power distribution method described in this invention, the following steps are taken: The power demand prediction set is subjected to power capacity constraint verification based on the cabinet power connection model, and evaluation and screening are performed in conjunction with power matching degree, power fluctuation, and power supply safety margin to form a selected set of power distribution schemes.

[0032] Based on the cabinet power connection model, power supply capacity constraint verification and power trimming are performed on the power demand prediction set to form a power allocation scheme for equipment.

[0033] Based on the equipment power supply allocation scheme and power state sequence, the power matching degree parameter, power fluctuation parameter and power supply safety margin parameter of the equipment power supply allocation scheme are obtained by statistical analysis, forming a power supply allocation evaluation parameter set;

[0034] Based on the power allocation evaluation parameter set, the power allocation schemes for equipment are comprehensively ranked in conjunction with the safety margin threshold. The power allocation schemes that meet the evaluation rules are selected to form the power allocation scheme selection set.

[0035] As a preferred embodiment of the AI-optimized cloud computer cabinet power distribution method described in this invention, the steps are as follows: The selected set of power distribution schemes is mapped to a set of power distribution parameters based on the port-path mapping relationship, and a power deviation assessment is combined with the execution state update to generate a cloud computer cabinet power distribution state set.

[0036] Based on the port-path mapping relationship in the cabinet power connection model, the selected set of power allocation schemes is mapped to the corresponding internal power management object of the device, and the set of power allocation parameters is obtained.

[0037] The actual power value of the power status sequence in the internal power management object of the device is compared with the allocated power value in the power allocation parameter set to obtain the power deviation value, and then integrated into a power deviation evaluation value set.

[0038] The power distribution parameter set whose deviation value exceeds the deviation threshold in the power supply deviation assessment value set is updated in status, and the corresponding update time stamp is written to generate the power distribution status set of the cloud computer cabinet.

[0039] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the AI-optimized cloud computer cabinet power distribution method as described in the first aspect of the present invention.

[0040] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the AI-optimized cloud computer cabinet power distribution method as described in the first aspect of the present invention.

[0041] The beneficial effects of this invention are as follows: by constructing a power demand prediction feature set and using a recurrent neural network model to perform forward inference, accurate prediction of the power demand of industrial cloud computing equipment is achieved, providing a data foundation for dynamic power allocation; by performing power supply capacity constraint verification and combining multi-dimensional evaluation and screening, the effect of ensuring power supply security and stability in complex industrial cloud computing environments is achieved, while improving energy efficiency while ensuring power supply security and stability. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of an AI-optimized power distribution method for cloud computer cabinets.

[0044] Figure 2 A flowchart for constructing the cabinet power connection model.

[0045] Figure 3 This is a flowchart for generating electrical power state sequences and predicting power.

[0046] Figure 4 A flowchart for generating and updating the power allocation scheme. Detailed Implementation

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides an AI-optimized cloud computer cabinet power distribution method, including the following steps:

[0051] S1. Based on the power supply topology information of the cloud computer cabinet, perform hierarchical analysis of the power supply interface, interface connection relationship and power supply path and establish hierarchical reference relationship to build the cabinet power supply connection model.

[0052] Based on the power supply topology information of the cloud computer cabinet, the power supply structure of the cloud computer cabinet is abstracted at the node level based on the power network topology model. Interface identifiers are assigned to interfaces with power supply start-point and power supply end-point attributes, and a power supply interface layer interface set is generated.

[0053] Furthermore, the set of power supply interface numbers for computing devices is extracted from the power supply topology information and mapped one by one to the node candidate set of the power network topology model. Based on the node directionality, current supply direction definition, and connection direction of the interface in the power supply topology model, the power supply start-point attribute interface and the power supply end-point attribute interface are determined. A unique interface identifier is assigned to each power supply start-point attribute interface and power supply end-point attribute interface (for example, the interface identifier consists of a cabinet identifier field, an interface type field, and a sequence number field). The interface identifier, interface attributes, and the positional relationship of the interface in the power supply topology are associated and stored to form a power supply interface layer interface set.

[0054] It should be noted that the power supply topology information of the cloud computer cabinet is obtained by organizing the structured configuration records formed during the planning and manufacturing stages of the cloud computer cabinet. This includes the set of power supply interface numbers for computing devices, interface type attributes, cable connection relationships between interfaces, and the physical location relationship between interfaces and the inside of the cabinet. The power network topology model is a structured model used to describe power supply nodes and the electrical connections between nodes. It abstracts power supply interfaces as nodes, the electrical connections between interfaces as connections, and organizes the power supply path by the connections. This model is used to depict the topology and reachability of electrical energy from the power supply start point to the power supply end point.

[0055] Using the power supply interface layer interface set as the node set in the power network topology model, and combining the power supply topology information of the cloud computer cabinet, the connection endpoints are paired and the connection attributes are recorded to generate a power supply connection layer connection relationship set.

[0056] Furthermore, using the power supply interface layer interface set as the node set in the power network topology model, the interface identifiers in the power supply interface layer interface set are matched one by one with the cable connection relationships between the interfaces in the power supply topology information of the cloud computer cabinet. The connection pairs between interface identifiers are determined according to the cable start interface number and end interface number, and the connection attributes of each connection pair (such as the cable type field and the cable rated load-bearing parameter field) are recorded. The interface identifiers and corresponding connection attributes are structured and organized to generate the power supply connection layer connection relationship set.

[0057] Based on the interface set of the power supply interface layer, the interface identifiers are linked in a chain according to the path reachability rules in the power network topology model along the connection order of the power supply connection layer connection relationship set to generate the path relationship set of the power supply path layer.

[0058] Furthermore, based on the power supply interface layer interface set, under the constraints of path reachability rules, the starting interface identifiers of all paths are determined, and the cable connection relationship between interfaces in the power supply topology information of the cloud computer cabinet is used as the unique connection constraint to perform a step-by-step traversal. In each traversal, only the interface identifiers that meet the node direction consistency and connectivity reachability conditions in the path reachability rules are associated. When the traversal reaches the interface identifier with the power supply endpoint attribute, the current traversal process is terminated, resulting in a corresponding ordered link. The ordered link is assigned a power supply path identifier, and the power supply path identifier is uniformly associated and stored with the corresponding ordered link to generate a power supply path layer path relationship set.

[0059] It should be noted that the path reachability rules are set based on the interface connection direction, cable connection relationship and power supply start-end constraint already determined in the power supply topology information of the cloud computer cabinet. By abstracting the power supply interface as a topology node and the cable connection as a connection edge between nodes, the reachability between nodes is limited to only when a physical connection exists and the direction is consistent.

[0060] Extract power supply path identifiers from the power supply path layer path relationship set, establish a corresponding relationship with interface identifiers, and generate port-path mapping relationship.

[0061] Furthermore, power supply path identifiers are extracted from the path relationship set of the power supply path layer, and a one-to-one correspondence between interface identifiers and power supply path identifiers is established based on the positional relationship of interface identifiers in the ordered link. When the same interface identifier appears only in one ordered link, the mapping and binding are completed directly, generating a port-path mapping relationship. Here, the port refers to the connection endpoint of the power supply interface in the power supply path, and the port and interface correspond one-to-one.

[0062] Based on the interface set of the power supply interface layer, the connection relationship set of the power supply connection layer, and the path relationship set of the power supply path layer, a hierarchical reference relationship between interface identifier, connection endpoint, and path link is established through hierarchical topology modeling to construct a cabinet power supply connection model.

[0063] Furthermore, using the interface identifiers in the power supply interface layer interface set as the smallest modeling unit, the interface identifiers are mapped to nodes in the power network topology model; using the cable start and end interface identifiers recorded in the power supply connection layer connection relationship set as connectivity constraints between nodes, a lower-level reference relationship between interface identifiers and connection endpoints is established; the ordered links recorded in the power supply path layer path relationship set are read, and the interface identifier-connection endpoint reference relationship is merged into the corresponding power supply path identifier level by level according to the link order, forming a hierarchical nested structure of interface identifier-connection endpoint-path link, where the interface identifier is the bottom-level node element, the connection endpoint is the inter-layer connectivity relationship, and the power supply path identifier is the upper-level aggregation index; the reference relationships of each level are uniformly organized and stored according to a fixed field order to form the cabinet power supply connection model.

[0064] S2. Extract the internal power management object of the device corresponding to the power supply endpoint from the cabinet power connection model, and perform synchronous sampling and power calculation to generate the power state sequence.

[0065] Extract interface identifiers with power supply endpoint attributes from the cabinet power connection model and map them to a set of power management objects inside the device.

[0066] Furthermore, interface identifiers with power supply endpoint attributes are filtered from the cabinet power connection model according to their hierarchical position in the power supply path layer. Using the power supply path identifier as an aggregation index, the power supply endpoint interface identifiers under the same path are uniformly mapped to the corresponding internal power management object number of the device (e.g., PMO-1). This transforms the power supply path relationship represented by the path relationship set of the power supply path layer into a set of internal power management objects of the device.

[0067] It should be noted that the device refers to the computing node device deployed in the cloud computing cabinet and connected to the cabinet power connection model through the power supply endpoint interface. The computing node device includes server nodes, storage computing nodes and network computing nodes.

[0068] Synchronous sampling is performed on the set of power management objects inside the device, and timestamp alignment, abnormal sampling point removal and sampling consistency verification are performed to generate power supply sampling data.

[0069] Furthermore, using the power supply path identifier corresponding to the path relationship set in the power supply path layer as the synchronization boundary, the internal power management objects in the device internal power management object set under the same path are sampled for voltage and current values ​​in the same period, and a unified time stamp is written for each sampled record. The sampling time stamps are rearranged and aligned in units of power supply path identifier, and only sampling records with completely consistent time stamps within the same power supply path are retained. Combined with the interface rated parameters, the voltage and current values ​​are judged for out-of-bounds and abnormal sampling points are removed (for example, when the current value exceeds the upper limit of the rated current of the corresponding interface, or the voltage value exceeds the voltage value range allowed by the interface, the corresponding record is deleted), forming power supply sampling data that has both path consistency and time consistency.

[0070] It should be noted that the interface rated parameters refer to the permissible working boundary parameters formed during the planning and manufacturing stage of the cloud computer cabinet, including the permissible voltage range, the upper limit of current, and the maximum permissible power value determined by the upper limit of voltage and current (e.g., the maximum permissible power value is 500 W).

[0071] Power values ​​are obtained from power supply sampling data and integrated into an electrical power state sequence.

[0072] Furthermore, using the device's internal power management object number and sampling time marker as a joint index, the voltage value, current value, and power factor (which are output in real time by the power monitoring chip corresponding to the device's internal power management object or determined by the device's nominal parameters) in each sampling record are multiplied to obtain the instantaneous power value at the corresponding moment. These values ​​are then connected in series according to time order to form an electrical power state sequence, which serves as the input basis for subsequent power change feature extraction and power demand prediction steps.

[0073] S3. Construct a power demand prediction feature set based on the power state sequence, and perform forward inference through the trained recurrent neural network model to generate a set of power demand prediction values.

[0074] The power state sequence is segmented, and the power change amplitude and power change trend features are extracted.

[0075] Furthermore, a sliding segmentation is performed on the power state sequence to obtain a set of power subsequences arranged in chronological order (e.g., five consecutive sampling time markers within a time window). The power values ​​of adjacent time markers in each power subsequence are calculated by difference and arranged in order to form a power difference value sequence. The maximum and mean absolute values ​​of the power difference value sequence represent the power change amplitude characteristics, while the sign consistency and cumulative sum direction of the power difference value sequence represent the power change trend characteristics. The power change trend characteristics and power change amplitude characteristics are then output.

[0076] The power change trend features and the power change trend features are combined and normalized to generate a power demand prediction feature set.

[0077] Furthermore, the maximum absolute value and mean absolute value of the power change amplitude feature are concatenated with the sign consistency, cumulative sum and direction of the power change trend feature in a fixed field order to form the current window feature vector. Based on the historical window feature vector set corresponding to the historical power state sequence, the minimum and maximum values ​​of each field are statistically obtained. The current window feature vector is then subjected to min-max normalization and clipped to a fixed value range, such as [0, 1], to generate a power demand prediction feature set.

[0078] Using historical power state sequences as training data, the parameters of the recurrent neural network model are iteratively trained by minimizing the mean square error loss function to obtain the trained recurrent neural network model.

[0079] Furthermore, the historical power state sequence is grouped according to the internal power management object number of the device, and the power value of the next time window corresponding to the feature vector of each window is used as the supervision label value to form the training sample. The training sample is divided into training batches, and the parameters of the recurrent neural network model are initialized. The recurrent neural network model adopts the input layer to receive the window feature vector sequence, the gated recurrent unit layer to update the hidden state in time order, and the fully connected layer to map the current hidden state to the predicted value of the next time window. The mean square error between the predicted value and the supervision label value in the training sample is obtained as the loss value. The error signal is gradually backpropagated from the last time step to the first time step through the loss function and time backpropagation to obtain the gradient of the weight matrix and the bias vector. The parameters are updated according to the learning rate (e.g., 0.001). The iteration rounds (e.g., 50 times) are repeated until the loss value is less than the convergence threshold and convergence is achieved, thus obtaining the trained recurrent neural network model.

[0080] It should be noted that the recurrent neural network model is a neural network model used to process time series data. By introducing hidden states between adjacent time steps and recursively updating them, the output of the current time step depends simultaneously on the current input features and the state information of historical time steps, thereby achieving the modeling and prediction of the time-series dependencies of the power state sequence. The convergence threshold is set based on the statistical fluctuation amplitude of the loss value of the training batch in adjacent rounds during the stable phase, and the exemplary value range is usually 10. -6 Up to 10 -3 .

[0081] The mean square of the prediction error and the convergence criterion expressions are as follows:

[0082] ;

[0083] in, The loss value is formed by the mean of the squared errors within the training batch; The number of training samples within a training batch; For the recurrent neural network model, the first The predicted power demand value for the next time window output from each training sample; For the first The actual power value of the next time window corresponding to each training sample; This is the index for the training samples.

[0084] The power state sequence is input into the trained recurrent neural network model, and the power demand for future time windows is predicted through forward inference, generating a set of predicted power demand values.

[0085] Furthermore, a window feature vector sequence is generated for the power state sequence according to the sliding time window. The window feature vector sequence is then input into the trained recurrent neural network model step by step. The model passes through the input layer, the gated recurrent unit layer for hidden state recursion, and the fully connected layer for regression to output the power demand prediction value for the next time window. The power demand prediction value for the next time window is aggregated according to the power management object number inside the device and written into the prediction time marker to generate a set of power demand prediction values ​​(e.g., the future time window length is 5 consecutive sampling time markers).

[0086] S4. Based on the cabinet power connection model, the power demand prediction set is checked for power capacity constraints, and the power matching degree, power fluctuation and power supply safety margin are evaluated and screened to form a selected set of power supply allocation schemes.

[0087] Based on the cabinet power connection model, power supply capacity constraint verification and power trimming are performed on the power demand prediction set to form a power allocation scheme for the equipment.

[0088] Furthermore, based on the port-path mapping relationship in the cabinet power connection model, the set of predicted power demand values ​​is associated with the power supply path identifier and the interface identifier. The allowed voltage range, current upper limit, and maximum allowable power value of the interface are read from the interface rated parameters. The predicted power demand values ​​of the same path are summarized with the power supply path identifier as the boundary and compared with the maximum allowable power value of the interface identifier at the starting point of the path. When the summarized predicted power demand values ​​exceed the maximum allowable power value, they are synchronously compressed according to the proportion of predicted power demand values ​​within the same path (for example, the synchronous compression coefficient is 0.83). The predicted power that still exceeds the maximum allowable power value after compression is truncated at the upper limit (for example, the predicted power value is replaced with the maximum allowable power value of the interface), and the allocated power supply value is written into the equipment power supply allocation scheme.

[0089] Based on the equipment power allocation scheme and power state sequence, the power matching degree parameter, power fluctuation parameter and power supply safety margin parameter of the equipment power allocation scheme are obtained by statistical analysis, forming a power allocation evaluation parameter set.

[0090] Furthermore, the allocated power supply value in the equipment power supply allocation scheme is paired window by window with the actual power value sequence in the power state sequence. The ratio of the absolute deviation between the allocated power supply value and the actual power value to the allocated power supply value is used as the power matching degree parameter. The dispersion statistic of the actual power value sequence within the time window is used as the power fluctuation parameter. The difference between the maximum allowable power value and the allocated power supply value is used as the power supply safety margin parameter. The equipment internal power management object number, prediction time stamp, power matching degree parameter, power fluctuation parameter, and power supply safety margin parameter are organized and stored to form a power supply allocation evaluation parameter set.

[0091] Based on the power allocation evaluation parameter set, the power allocation schemes for equipment are comprehensively ranked in conjunction with the safety margin threshold. The power allocation schemes that meet the evaluation rules are selected to form the power allocation scheme selection set.

[0092] Furthermore, based on the interface rated parameters in the cabinet power connection model, the power allocation evaluation parameter set is associated with the power allocation scheme of the equipment according to the internal power management object number and the prediction time stamp. Allocation schemes with power safety margin parameters less than the safety margin threshold are eliminated. Then, the remaining allocation schemes are sorted and the first few items (e.g., the first 5 items) of the sorted sequence are extracted to form the power allocation scheme selection set.

[0093] It should be noted that the safety margin threshold is set based on the maximum allowable power value in the rack design configuration data. The difference sequence between the peak power and the maximum allowable power value is extracted from the historical power state sequence, and a fixed percentile (e.g., the 95th percentile) of the difference sequence is taken as the safety margin threshold. An exemplary safety margin threshold range is 0W to 100W.

[0094] S5. Based on the port-path mapping relationship, the selected power allocation scheme is mapped to the power allocation parameter set, and combined with the power deviation assessment to perform status updates, a power allocation status set for the cloud computer cabinet is generated.

[0095] Based on the port-path mapping relationship in the cabinet power connection model, the selected set of power allocation schemes is mapped to the corresponding internal power management objects of the equipment, and the set of power allocation parameters is obtained.

[0096] Furthermore, the interface identifier and power supply path identifier are retrieved from the port-path mapping relationship in the cabinet power connection model, and a mapping check code is generated. The mapping check code is obtained by concatenating the interface identifier and power supply path identifier in a fixed order and calculating the hash value. Based on the selected set of power allocation schemes, the safety margin value between the allocated power value and the upper limit of the power limit is obtained one by one. When the safety margin value is less than the safety margin threshold, a truncation mark is written. The internal power management object number, prediction time mark, allocated power value, upper limit of power limit, interface identifier, power supply path identifier, mapping check code, and truncation mark of the equipment are organized and stored to form a power allocation parameter set.

[0097] The actual power value of the power state sequence in the internal power management object of the device is compared with the allocated power value in the power allocation parameter set to obtain the power deviation value, and then integrated into a power deviation evaluation value set.

[0098] Furthermore, using the power distribution parameter set as the alignment benchmark, the power status sequence is matched according to the internal power management object number of the equipment, and the predicted time marker is mapped to the starting sampling time marker of the corresponding predicted time window. The actual power value of the corresponding sampling time marker is read and the difference operation is performed with the allocated power supply power value to obtain the power deviation value. The absolute mean and absolute maximum values ​​of the power deviation value are taken with the predicted time marker as the statistical boundary and statistically formed into a power supply deviation evaluation value set.

[0099] The power distribution parameter set whose deviation value exceeds the deviation threshold in the power supply deviation assessment value set is updated in status, and the corresponding update time stamp is written to generate the power distribution status set of the cloud computer cabinet.

[0100] Furthermore, based on the power supply deviation assessment value set, the power supply allocation parameter set record is located one by one according to the internal power management object number and the prediction time stamp, and the mapping check code is verified. When the mapping check code is consistent, the power deviation value is read and the absolute value of the power deviation value is obtained. When the absolute value of the power deviation value exceeds the deviation threshold, the status field of the power supply allocation parameter set record is updated to the "deviation exceeds deviation threshold" flag and the update time stamp is written synchronously. All updated power supply allocation parameter set records are summarized to generate the power allocation status set of the cloud computer cabinet.

[0101] It should be noted that the deviation threshold is set based on the statistical distribution of the absolute value of the power deviation during the stable operation phase of the historical power state sequence. It is obtained by taking the fixed percentile (95th percentile) of the absolute value of the power deviation and truncating it. An exemplary value range is 10W to 100W.

[0102] This embodiment also provides a computer device applicable to the AI-optimized cloud computer cabinet power distribution method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the AI-optimized cloud computer cabinet power distribution method proposed in the above embodiment.

[0103] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0104] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the AI-optimized cloud computer cabinet power distribution method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0105] In summary, this invention achieves accurate prediction of power demand for industrial cloud computing equipment by constructing a power demand prediction feature set and using a recurrent neural network model to perform forward inference, thus providing a data foundation for dynamic power allocation. By performing power supply capacity constraint verification and combining multi-dimensional evaluation and screening, it achieves the effect of ensuring power supply security and stability in complex industrial cloud computing environments, while improving energy efficiency while ensuring power supply security and stability.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cloud computer cabinet power distribution method based on AI optimization, characterized in that: include, Based on the power supply topology information of the cloud computer cabinet, the power supply interface, interface connection relationship and power supply path are analyzed in layers and a hierarchical reference relationship is established to construct the cabinet power supply connection model. Extract the internal power management object of the device corresponding to the power supply endpoint from the cabinet power connection model, and perform synchronous sampling and power calculation to generate an electrical power state sequence; A power demand prediction feature set is constructed based on the power state sequence, and forward inference is performed through a trained recurrent neural network model to generate a set of power demand prediction values. Based on the cabinet power connection model, the power demand prediction set is checked for power capacity constraints, and the power matching degree, power fluctuation and power supply safety margin are evaluated and screened to form a set of selected power allocation schemes. Based on the port-path mapping relationship, the selected set of power allocation schemes is mapped to the set of power allocation parameters, and combined with the power deviation assessment to perform status updates, a set of power allocation statuses for the cloud computer cabinet is generated.

2. The AI-optimized cloud computer cabinet power distribution method as described in claim 1, characterized in that: The process of performing layered analysis of power supply interfaces, interface connection relationships, and power supply paths based on the power supply topology information of the cloud computer cabinet is as follows: Based on the power supply topology information of the cloud computer cabinet, the power supply structure of the cloud computer cabinet is abstracted at the node level based on the power network topology model. Interface identifiers are assigned to interfaces with power supply start-point and power supply end-point attributes, and a power supply interface layer interface set is generated. Using the power supply interface layer interface set as the node set in the power network topology model, and combining the power supply topology information of the cloud computer cabinet, the connection endpoints are paired and the connection attributes are recorded to generate a power supply connection layer connection relationship set. Based on the interface set of the power supply interface layer, the interface identifiers are linked in a chain according to the path reachability rules in the power network topology model along the connection order of the power supply connection layer connection relationship set to generate the path relationship set of the power supply path layer.

3. The AI-optimized cloud computer cabinet power distribution method as described in claim 2, characterized in that: The steps for establishing hierarchical reference relationships and constructing the cabinet power connection model are as follows. Extract power supply path identifiers from the power supply path layer path relationship set, establish a corresponding relationship with interface identifiers, and generate port-path mapping relationship; Based on the interface set of the power supply interface layer, the connection relationship set of the power supply connection layer, and the path relationship set of the power supply path layer, a hierarchical reference relationship between interface identifier, connection endpoint, and path link is established through hierarchical topology modeling to construct a cabinet power supply connection model.

4. The AI-optimized cloud computer cabinet power distribution method as described in claim 3, characterized in that: The steps are as follows: extract the internal power management object of the device corresponding to the power supply endpoint from the cabinet power connection model, perform synchronous sampling and power calculation, and generate an electrical power state sequence. Extract interface identifiers with power supply endpoint attributes from the cabinet power connection model and map them to a set of power management objects inside the device. Synchronous sampling is performed on the set of power management objects inside the device, and timestamp alignment, abnormal sampling point removal and sampling consistency verification are performed to generate power supply sampling data; Power values ​​are obtained from power supply sampling data and integrated into an electrical power state sequence.

5. The AI-optimized cloud computer cabinet power distribution method as described in claim 4, characterized in that: The steps for constructing a power demand prediction feature set based on the electrical power state sequence are as follows: The power state sequence is segmented, and the power change amplitude and power change trend features are extracted. The feature combination processing is performed on the change magnitude feature and the change trend feature, and then normalization processing is performed to generate a power demand prediction feature set.

6. The AI-optimized cloud computer cabinet power distribution method as described in claim 5, characterized in that: The process of generating a set of predicted power demand values ​​by performing forward inference using a trained recurrent neural network model is as follows: Using historical power state sequences as training data, the parameters of the recurrent neural network model are iteratively trained by minimizing the mean square error loss function to obtain the trained recurrent neural network model. The power state sequence is input into the trained recurrent neural network model, and the power demand for future time windows is predicted through forward inference, generating a set of predicted power demand values.

7. The AI-optimized cloud computer cabinet power distribution method as described in claim 6, characterized in that: The process involves performing power capacity constraint verification on the power demand prediction set based on the cabinet power connection model, and then evaluating and filtering based on power matching degree, power fluctuation, and power supply safety margin to form a selected set of power allocation schemes. The steps are as follows. Based on the cabinet power connection model, power supply capacity constraint verification and power trimming are performed on the power demand prediction set to form a power allocation scheme for equipment. Based on the equipment power supply allocation scheme and power state sequence, the power matching degree parameter, power fluctuation parameter and power supply safety margin parameter of the equipment power supply allocation scheme are obtained by statistical analysis, forming a power supply allocation evaluation parameter set; Based on the power allocation evaluation parameter set, the power allocation schemes for equipment are comprehensively ranked in conjunction with the safety margin threshold. The power allocation schemes that meet the evaluation rules are selected to form the power allocation scheme selection set.

8. The AI-optimized cloud computer cabinet power distribution method as described in claim 7, characterized in that: The process involves mapping the selected power allocation scheme set based on the port-path mapping relationship to a power allocation parameter set, and combining this with power deviation assessment to perform status updates, thereby generating a cloud computer cabinet power allocation status set. The steps are as follows: Based on the port-path mapping relationship in the cabinet power connection model, the selected set of power allocation schemes is mapped to the corresponding internal power management object of the device, and the set of power allocation parameters is obtained. The actual power value of the power status sequence in the internal power management object of the device is compared with the allocated power value in the power allocation parameter set to obtain the power deviation value, and then integrated into a power deviation evaluation value set. The power distribution parameter set whose deviation value exceeds the deviation threshold in the power supply deviation assessment value set is updated in status, and the corresponding update time stamp is written to generate the power distribution status set of the cloud computer cabinet.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the AI-optimized cloud computer cabinet power distribution method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the AI-optimized cloud computer cabinet power distribution method according to any one of claims 1 to 8.

Images